computing pixel error with mask

[xiuliren]

Sachin found that current pixel error computation do not consider the mask:
https://github.com/seung-lab/znn-release/blob/master/python/train.py#L98

will fix this later.

[nicholasturner1]

This has been fixed in the dev branch

[xiuliren]

@nicholasturner1 I did not find the corresponding code.
https://github.com/seung-lab/znn-release/blob/development/python/train.py#L135

[nicholasturner1]

Hmm, ok. I must have only fixed this in the sparse training branch. I can push a fix for this today.